sensors

Article
A Quantile Mapping Bias Correction Method Based
on Hydroclimatic Classification of the Guiana Shield
Justine Ringard 1, *, Frederique Seyler 2 and Laurent Linguet 1
1 Université de Guyane, UMR ESPACE-DEV, IRD, Université de La Réunion, Université de Montpellier,
F-97300 Cayenne, French Guiana; linguetlaur@gmail.com
2 IRD, UMR ESPACE-DEV, Université de Guyane, Université de La Réunion, Université de Montpellier,
Maison de la Télédétection, 500 rue Jean-François Breton, F-34093 Montpellier CEDEX 5, France;
frederique.seyler@ird.fr
* Correspondence: justine.ringard@ird.fr; Tel.: +594-(0)-594-299-273

Academic Editor: Assefa M. Melesse

Received: 28 March 2017; Accepted: 12 June 2017; Published: 16 June 2017

Abstract: Satellite precipitation products (SPPs) provide alternative precipitation data for regions with
sparse rain gauge measurements. However, SPPs are subject to different types of error that need
correction. Most SPP bias correction methods use the statistical properties of the rain gauge data to
adjust the corresponding SPP data. The statistical adjustment does not make it possible to correct the
pixels of SPP data for which there is no rain gauge data. The solution proposed in this article is to correct
the daily SPP data for the Guiana Shield using a novel two set approach, without taking into account
the daily gauge data of the pixel to be corrected, but the daily gauge data from surrounding pixels.
In this case, a spatial analysis must be involved. The first step defines hydroclimatic areas using a
spatial classification that considers precipitation data with the same temporal distributions. The second
step uses the Quantile Mapping bias correction method to correct the daily SPP data contained within
each hydroclimatic area. We validate the results by comparing the corrected SPP data and daily rain
gauge measurements using relative RMSE and relative bias statistical errors. The results show that
analysis scale variation reduces rBIAS and rRMSE significantly. The spatial classification avoids mixing
rainfall data with different temporal characteristics in each hydroclimatic area, and the defined bias
correction parameters are more realistic and appropriate. This study demonstrates that hydroclimatic
classification is relevant for implementing bias correction methods at the local scale.

Keywords: quantile mapping bias correction; hydroclimatic area; temporal distribution;

TRMM-TMPA 3B42V7; Guiana Shield

1. Introduction
Given the current context of climate change, it is essential to improve our understanding of the
spatial and temporal dynamics of precipitation on the global and regional scales. Precipitation is an
essential part of the global water cycle and its measurement is particularly crucial. Observation
of precipitation at high spatial resolution is very important for monitoring and forecasting
extreme weather events such as floods and droughts, but also for obtaining input data for
hydrological applications and climate studies. However, obtaining accurate gauge-based precipitation
measurements at high spatial resolution is difficult because of various technical and practical
limitations [1]. Technical limitations can be exposed during heavy rainfall when water can accumulate
at a rate faster than can be cleared by the calibration trough of the measurement device. Conversely,
during light rainfall, water can evaporate from the collector. Practical limitations arise from the
challenges associated with the installation and maintenance of a dense network of measurement
devices in areas that are difficult to access, such as mountains, deserts, and primary forests [1,2].

Sensors 2017, 17, 1413; doi:10.3390/s17061413 www.mdpi.com/journal/sensors

Sensors 2017, 17, 1413 2 of 17

Satellite precipitation products (SPPs) such as the Tropical Rainfall Measuring Mission
Multisatellite Precipitation Analysis (TRMM-TMPA 3B42) [3–5], Climate Prediction Center MORPHing
(CMORPH) product [6], and Precipitation Estimation from Remotely Sensed Information using
Artificial Neural Networks (PERSIANN) [7] provide alternatives for obtaining precipitation data
in regions with an insufficient distribution of rain gauge measurement stations. These SPPs with high
spatial resolution (0.25◦ × 0.25◦ ) and high temporal resolution (3 h) [8–15] are produced using data
obtained by different instruments deployed on several satellites. The algorithms of these products
link cloud brightness temperature, measured in the infrared band, with data relating to the size and
characteristics of hydrometeors, as measured by microwaves [1,14,16].
All the above blended rainfall products are subject to different types of error depending on the
quality of the measurements made by the sensor as well as the climate, topography, season, and
local climatic regime. Several studies have addressed SPP failures in different geographic regions
based on the intensity of precipitation and/or seasonal precipitation. The first evident defect of
SPPs is that they overestimate low daily intensities (<2 mm/day) and underestimate high daily
intensities (>20 mm/day) [8,11–14,17–19]. A second defect, observed by many authors, is that SPP
performance depends on season, with greater errors detected in winter compared with summer or
rainy seasons [9,10,12,18,20–25]. Ebert et al. [24] showed that as the precipitation regime tends toward
deep convection, the accuracy of the satellite estimates improves. A third defect observed in SPPs is
that their effectiveness depends on the climatic regime and that they show poorer performance in the
driest areas [12,13,17,26]. A fourth defect relates to the difference in performance of SPPs according
to topography, with low efficiency reported for SPPs in mountainous regions [9,17,27–29]. Finally,
SPPs have a low ability to detect the daily precipitated volume [27,30,31]. Indeed, SPPs are less able to
detect daily rainfall volume than rainfall occurrence. To correct all these defects, bias adjustment is
essential prior to the use of SPPs in hydrological applications.
Different bias adjustment approaches are used to improve the quality of SPP data. The linear
correction method corrects the average precipitation value based on the differences between the rain
gauge data and satellite data. However, this method does not correct the variance and all events are
adjusted with the same correction factor [15,32–37]. The Local Intensity Scaling method combines
a precipitation threshold with linear scaling [32,37,38]. This method separately corrects wet-day
frequency and wet-day intensity, applied pointwise and individually for each day of the year, and the
estimated precipitation is corrected using a scaling factor. Nevertheless, the results obtained with this
method are limited because, as with linear correction, the standard deviation is not corrected and all
events are adjusted using the same correction factor. The Power Transformation method corrects the
mean and variance of the temporal series of estimated precipitation [37,39–41]. This is a nonlinear
correction in an exponential form that combines the correction of the coefficient of variation with linear
scaling. The coefficient of variation of both daily and multiple-day precipitation amounts depends on
the wet-day frequency, but this correction does not adjust the frequency of wet days [40]. The Quantile
Mapping method (QM) [32,37,39,42–52], also named Distribution Mapping or the Quantile–Quantile
method, adjusts the cumulative distribution of estimated data to the cumulative distribution of rain
gauge data using a transfer function. This correction can capture the evolution of the mean and the
variability of precipitation while matching all statistical moments. Most of these methods use rain
gauge data to correct SPP data located in the same location (with respect to the satellite pixel). The aim
of this article is to correct the SPP data for the Guiana Shield using a novel two set approach, pixel by
pixel, without taking into account the daily rain gauge data of the pixel to be corrected, but instead
using the daily rain gauge data of the surrounding pixels. We introduce the concept of spatial scale
change analysis. The concept of scale must be involved in any spatial statistical analysis.
All the above bias correction methods are used to correct SPP estimates to provide results that are
acceptable on the global scale; however, these methods are limited when correcting SPPs on the local
scale. One limitation of applying these methods on the global scale is that precipitation estimates are
corrected using the same scaling factor or the same correction coefficients without consideration of the
Sensors 2017, 17, 1413 3 of 17

Sensors 2017,
disparities 17, 1413 the series to be corrected. Another weakness is that the statistical profile
between 3 of of
17 the

precipitation series is considered without accounting for the temporal profile. All these difficulties
are corrected using the same scaling factor or the same correction coefficients without consideration
limitofthe correction of SPPs on the local scale.
the disparities between the series to be corrected. Another weakness is that the statistical profile of
Here, we propose
the precipitation seriestois correct
considereddaily SPP accounting
without estimates for(TRMM-TMPA 3B42V7)
the temporal profile. usingdifficulties
All these a novel two
set approach. The first step defines hydroclimatic
limit the correction of SPPs on the local scale. areas using a spatial classification that considers
precipitation
Here,data with thetosame
we propose temporal
correct daily SPP distributions. The second 3B42V7)
estimates (TRMM-TMPA step usesusing
the Quantile
a novel twoMapping
set
biasapproach.
correctionThe method to correct
first step definesthe daily SPP data
hydroclimatic areascontained within
using a spatial each hydroclimatic
classification area, by
that considers
precipitation
defining data with
a calibration the same
set and temporalset.
a validation distributions. The second step uses the Quantile Mapping
bias correction
To identify method
the to correct
influence of thethe daily SPP
analysis data
scale oncontained withinresults,
the statistical each hydroclimatic area, bywere
three simulations
definingfor
conducted a calibration
the Guiana setShield
and a validation set. rain gauges. The first simulation was performed for
using 93 daily
the entireTostudy
identify
area,thei.e.,
influence
withoutofanythe hydroclimatic
analysis scale on the statistical
division. results,simulation
The second three simulations were
was undertaken
conducted for the Guiana Shield using 93 daily rain gauges. The first simulation was performed for
with the study area divided into six hydroclimatic areas. The third simulation was performed with
the entire study area, i.e., without any hydroclimatic division. The second simulation was undertaken
23 hydroclimatic areas. For each simulation, we compared both the daily precipitation estimated from
with the study area divided into six hydroclimatic areas. The third simulation was performed with
TRMM-TMPA 3B42V7 and the daily rain gauge measurements; and the precipitation corrected using
23 hydroclimatic areas. For each simulation, we compared both the daily precipitation estimated from
the QM method and
TRMM-TMPA dailyand
3B42V7 rainthe
gauge
dailymeasurements. We then validated
rain gauge measurements; the accuracy
and the precipitation of the correction
corrected using
basedtheon
QM the RMSEand
method anddaily
bias.rain gauge measurements. We then validated the accuracy of the correction
The remainder
based on the RMSE ofand
thisbias.
article is structured as follows: Section 2 describes the study area and
presents The
the data used. Section
remainder of this article 3 provides an overview
is structured of theSection
as follows: methodology
2 describesused
thetostudy
correct theand
area biases.
The presents
results arethepresented and discussed
data used. Section 3 providesin an
Section 4. Finally,
overview the conclusions
of the methodology usedare drawnthe
to correct in Section
biases. 5.
The results are presented and discussed in Section 4. Finally, the conclusions are drawn in Section 5.
2. Data
2. Data
2.1. Study Area
2.1. Study Area
The Guiana Plateau, also called the Guiana Shield, is a region in South America located north of
the Amazon The Guiana
River andPlateau, also
east of thecalled the Guiana
Orinoco Shield,
River. This areais aisregion
over 2inmillion
South America locatedsix
km2 . It spans north of
countries:
the Amazon River and east of the Orinoco River. This area is over 2 million
Colombia, Venezuela, Guyana, Suriname, French Guiana, and northern Brazil (Amapá, Roraima, and km 2. It spans six countries:

Colombia,
Pará). The Guiana Venezuela,
ShieldGuyana,
has poor Suriname,
soil, an French Guiana,
extensive riverand northern
system, andBrazil
dense (Amapá,
primary Roraima, and[53].
rainforest
Pará). The Guiana Shield has poor soil, an extensive river system, and dense primary rainforest [53].
This area accounts for 13% of the surface of the South American continent. In this study, the area
This area accounts for 13% of the surface of the South American continent. In this study, the area
considered lies within 2◦ S–6◦ N, 45◦ –62◦ W (Figure 1). The Guiana Shield is a region with high spatial
considered lies within 2° S–6° N, 45°–62° W (Figure 1). The Guiana Shield is a region with high spatial
variability
variabilityin precipitation [11,54].
in precipitation [11,54].The
Theannual
annual average differenceininprecipitation
average difference precipitation between
between the the littoral
littoral
zonezone
andand inland areas
inland cancan
areas reach
reach 2300
2300mm/year
mm/year [11].
[11]. ItItisisan
anarea
areasubject
subjectto to intense
intense andand local
local convective
convective
precipitation. The network of precipitation measurements is very sparse and located
precipitation. The network of precipitation measurements is very sparse and located primarily on the primarily on the
coastcoast
andand along thethe
along rivers. Most
rivers. Mostofofthe
thepresent
present population
population lives liveson
onthe
thecoast
coast
or or along
along thethe rivers
rivers and and
thus,thus,
is highly vulnerable
is highly vulnerableto to
flooding.
flooding.

Figure
Figure 1. Elevation
1. Elevation mapmap of Guiana
of the the Guiana Shield.
Shield. TheThe SRTM30
SRTM30 (Shuttle
(Shuttle Radar
Radar Topography
Topography Mission)
Mission) digital
digital elevation model is available at http://www.diva-gis.org/gdata. Dots represent the daily rain
elevation model is available at http://www.diva-gis.org/gdata. Dots represent the daily rain gauges
gauges available in French Guiana and northern Brazil.
available in French Guiana and northern Brazil.
Sensors 2017, 17, 1413 4 of 17

2.2. Rain Gauges

The daily rain gauge data used in this study come from 93 daily rain gauges distributed between
French Guiana and northern Brazil (Figure 1), including 18 in the Guyanese territory. North Brazilian
data come from the Brazil National Water Agency (ANA). They are freely available online [55].
Rainfall data from French Guiana are provided by Météo France. To have the most complete time
series and compare the SPP data, we use the period 2001–2012. The data are checked for quality and
used in the following analysis.

2.3. Precipitation Product

A recent study by Ringard et al. [11] compared different SPPs in the Guiana Shield (TRMM-TMPA
3B42RT, TRMM-TMPA 3B42V7, PERSIANN, and CMORPH). The results obtained show that for areas
with intense convective precipitation, TRMM-TMPA 3B42V7 performs better than the other products,
especially in the estimation of extreme precipitation events. In regions along the Amazon, the use of
PERSIANN is better. Finally, in the driest areas, TRMM-TMPA 3B42V7 and PERSIANN exhibit the
same level of performance.
The daily SPPs used in this study come from the TRMM TMPA 3B42 algorithm, which was
developed by NASA. A brief description of the TRMM-TMPA 3B42V7 product is given below.
TRMM TMPA 3B42 is a rainfall estimation product from the TRMM mission that combines satellite
and ground data [3–5,22,56,57]. The main data sources for TRMM-TMPA 3B42V7 are infrared GOES-W
(Geostationary Operational Environmental Satellite-West), GOES-E (East), GMS (Geostationary
Meteorological Satellite), Meteosat-5, Meteosat-7, and NOAA-12 geostationary satellites, as well as the
passive microwave radiometers of the TMI/TRMM (TRMM Microwave Imager), SSMI/DMSP (Special
Sensor Microwave Imager/Defense Meteorological Satellite Program), AMSU/NOAA (Advanced
Microwave Sounding Unit/National Oceanic and Atmospheric Administration) and AMSR-E/Aqua
(Advanced Microwave Scanning Radiometer-EOS) low orbit satellites.
During the last ten years, the TMPA algorithm has undergone three important updates to
incorporate data from new sensors into the algorithm [22]. TRMM-TMPA 3B42V7, the research
version, is available approximately two months after observation. The 3B42 algorithm runs in four
stages [13]: (1) passive microwave precipitation estimates are calibrated and combined, (2) infrared
precipitation estimates are generated using the calibrated data from the passive microwave sensors,
(3) the infrared and passive microwave data are combined, and (4) the data are rescheduled on a
monthly basis using the rain gauge data. Furthermore, TRMM-TMPA 3B42 uses precipitation estimates
directly from the passive microwave data when available, but it inserts infrared data when the passive
microwave data are unavailable [10]. TRMM-TMPA 3B42V7 algorithm output data have a 3 h time
resolution with rainfall amounts expressed in mm/h. The geographical area covered extends from
latitude 50◦ N–50◦ S for 3B42V7 with a 0.25◦ × 0.25◦ spatial grid resolution. The TRMM-TMPA 3B42V7
product data have been available since January 1, 1998 (through to the present day) [58].

3. Methods
We propose to correct daily TRMM-TMPA 3B42V7 estimates using a two set approach without
taking into account the daily rain gauge data of the pixel to be corrected, but instead using the
daily rain gauge data of the surrounding pixels. In this case, a spatial analysis must be involved.
This novel set approach is divided into two steps: (1) several scales are defined through the definition
of hydroclimatic zones; (2) bias correction using the Quantile Mapping approach is parameterized
with the daily data contained in each hydroclimatic area, through the definition of a calibration set and
a validation set.
Sensors 2017, 17, 1413 5 of 17

3.1. Definition of Hydroclimatic Area

The aim is to identify the influence of the analysis scale on the efficiency of the Quantile Mapping
correction method.
Sensors 2017, 17, 1413 For this purpose, we define hydroclimatic areas obtained from the rain gauge 5 of 17
precipitation series based on long-term monthly means of the rain gauge data. The hydroclimatic
areas are constructed using a hierarchical ascendant classification (HAC) [59,60]. The purpose of the
classification approach is to obtain groups of rainfall time series with similar profiles. The creation of
precipitation series into
such groups makes it possible to distribute the precipitation into different
different hydroclimatic areas.
At the
the initial
initialstage
stageofofthe
themethod,
method,each each rain gauge forms a
rain gauge forms a class, which class, which
makesmakes 93 classes.
93 classes. The
The method proceeds by reducing the number of classes. At each step, two
method proceeds by reducing the number of classes. At each step, two classes are grouped; thus, classes are grouped; thus,
reducing the number of classes. The two classes chosen for grouping are those whose dissimilarity is
the weakest; this dissimilarity value is called the aggregation index. Here, the aggregation index uses
the centers of gravity of the classes, as in the Ward Ward method
method [59].
[59].
Three simulations
simulationshave havebeen
beenimplemented
implemented totoaccount
account forfor
thethe
influence of scaling
influence analysis.
of scaling The first
analysis. The
simulation was performed on a single area, representing all the 93 rain gauges
first simulation was performed on a single area, representing all the 93 rain gauges (i.e., the entire(i.e., the entire study
area). area).
study The second simulation
The second was performed
simulation was performedon sixon classes representing
six classes six hydroclimatic
representing areas.
six hydroclimatic
The third simulation was undertaken on 23 classes representing 23 hydroclimatic
areas. The third simulation was undertaken on 23 classes representing 23 hydroclimatic areas. The areas. The second
simulation,
second carriedcarried
simulation, out onoutsix on
hydroclimatic areas,areas,
six hydroclimatic is theiscontinuation
the continuation of the work
of the work developed
developed in
Ringard
in Ringardet al. [11],
et al. who
[11], whoperform
perform a regional
a regionalanalysis
analysis ofofSPPs
SPPsininthese
thesesixsixareas.
areas. Figure
Figure 22 shows an
example of the classification
classification performed in this study for zone 6 only. In the exampleshown
performed in this study for zone 6 only. In the example shownininFigure
Figure 3,
three
3, threeareas
areasareare
created
createdfrom zone
from 6: Z6c11,
zone Z6c2,
6: Z6c11, Z6c2,and Z6b.
and Z6b.The 23 23
The hydroclimatic
hydroclimatic areas
areas areare
obtained
obtainedby
following
by followingthisthis
same scheme.
same scheme.

Figure 2. Diagram
Diagram of of the
the hierarchical ascendant classification for zone 6. 6. Red
Red dots
dots indicate the
dissimilarity values of the hydroclimatic groups conserved. Black squares in solid lines are the names
Dotted black
of the areas at different hierarchical levels. Dotted black squares
squares are
are classes
classes with
with only
only one
one rain
rain gauge,
gauge,
which are therefore unusable.
unusable.

3.2. Principles and Implementation of the Quantile Mapping (QM) Method

The second step uses the Quantile Mapping bias correction method to correct the daily SPP data
contained within each hydroclimatic area, by defining a calibration set and a validation set. The QM
method adjusts the distribution of daily satellite precipitation (Pss) with the distribution of daily rain
gauge precipitation (Poo) using
using aa transfer
transfer function
function (h).
(h). Figure
Figure 33presents
presents aaschematic
schematic of
ofthe
theQM
QMmethod.
method.
The transformation can be formulated as below [61,62]:
[61,62]:
( ) (1)
Po = h( Ps ) (1)
Sensors 2017, 17, 1413 6 of 17
Sensors 2017, 17, 1413 6 of 17
Sensors 2017, 17, 1413 6 of 17

Figure3.3.Schematic
Figure Schematicofofthe
theQuantile
QuantileMapping
Mappingmethod.
method.The
Thedistribution
distributionfunction
functionof
ofthe
theSPP
SPPdata
dataisis
Figure
shifted 3.tothe
shiftedto Schematic of thefunction
thedistribution
distribution Quantile
functionofofMapping
therain
the method.
raingauge
gauge The distribution function of the SPP data is
data.
data.
shifted to the distribution function of the rain gauge data.
If the variable of interest has a known distribution, the transformation is defined as:
If the variable of interest has a known distribution, the transformation is defined as:
If the variable of interest has a known distribution, the transformation is defined as:
( ) (2)
Po = Fo−1 ( Fs ((Ps ))
) (2)
(2)
where Fs is the Cumulative Distribution Function (CDF) of Ps and is the inverse CDF of Po.
where Fs is the areCumulative Distribution Function related
(CDF) of P s and −1 is theforinverse CDFthe of quantile-
Po.
whereThere
Fs is the several statistical
Cumulative transformations
Distribution Function (CDF) of to P
the QM method modeling
s and Fo is the inverse CDF of Po .
There
quantile are several statistical transformations related to the QM method for modeling the quantile-
Thererelationship
are several[62]. The distribution-derived
statistical transformations relatedtransformation
to the QMusesmethod
a theoretical distribution
for modeling theto
quantile
solve relationship
Equation (2). [62]. The distribution-derived
Parametric transformations transformation
are used directlyuses
to a theoretical
model the distribution to
quantile-quantile
quantile-quantile relationship [62]. The distribution-derived transformation uses a theoretical
solve Equation
relationship (2). Parametric transformations are used directly to model the quantile-quantile
distribution to(Equation (1)). Finally,
solve Equation instead of assuming
(2). Parametric parametric
transformations distributions,
are used directly nonparametric
to model the
relationship
transformations (Equation
use (1)).
empirical Finally,
CDFs instead
to solve of assuming
Equation (2) parametric
or distributions,
nonparametric regressions nonparametric
such as cubic
quantile-quantile relationship (Equation (1)). Finally, instead of assuming parametric distributions,
transformations
smoothing useto
splines empirical
solve CDFs to(1).
Equation solve Equation
Several (2) or nonparametric
approaches are possible; regressions
here, a such as
smoothing cubicis
spline
nonparametric transformations use empirical CDFs to solve Equation (2) or nonparametric regressions
smoothing splines to solve Equation (1). Several approaches are possible; here, a smoothing spline is
such as cubic smoothing splines to solve Equation (1). Several approaches are possible; here, on
used to fit the quantile-quantile plot of daily observed and daily modeled time series. We relied a
used
the to
work fit the
of quantile-quantile
Gudmundsson et plot
al. [62] of
todaily observed
implement this and daily
method. modeled time series. We relied on
smoothing spline is used to fit the quantile-quantile plot of daily observed and daily modeled time
the work of Gudmundsson et al. [62] to implement this method.
series. We relied on the work of Gudmundsson et al. [62] to implement this method.
Calibration of the QM Method and Correction of SPP Time Series
Calibration
Calibration of the
of the QM
QM Method
Method and and Correction
Correction of of SPP
SPP Time
TimeSeries
Series
To implement the QM correction method we divided each hydroclimatic area into two parts:
To implement
calibration set and the QM
validation correction
set. We method
take the we
example
To implement the QM correction method we divided each hydroclimaticdivided
of a each hydroclimatic
hydroclimatic area
area into
area composed two
into oftwo parts:
four rain
parts:
calibration set
gauges represented
calibration and validation set.
by four pixels
set and validation We
set. We take
(Figure the
take the example of a hydroclimatic area composed
4).example of a hydroclimatic area composed of four rain of four rain
gauges represented by four pixels
gauges represented by four pixels (Figure 4). (Figure 4).

Figure 4. Diagram showing the calibration set (green) and validation set (red) for a hydroclimatic
Figure
Figureof
zone 4.four
4. Diagram showing
dailyshowing
Diagram rain thecalibration
gauges
the calibration
with setset
the TRMM-TMPA(green)
(green) andand validation
3B42V7 set (red)
grid.set (red)
validation for afor a hydroclimatic
hydroclimatic zone
zone of four daily rain gauges with the TRMM-TMPA
of four daily rain gauges with the TRMM-TMPA 3B42V7 grid. 3B42V7 grid.
The calibration set is shown in green and the validation set in red. The calibration process uses
The calibration
the calibration set toset is shown
adjust in green and
the distribution thedaily
of the validation set to
SPP data in match
red. The
thecalibration process
distribution of the uses
daily
the calibration
rain gauge dataset[62].
to adjust the distribution
A function of thethe
is set to adjust daily SPP data todata
precipitation match theadistribution
from of the daily
rainfall threshold set at
rain gauge data [62]. A function is set to adjust the precipitation data from a rainfall threshold set at
Sensors 2017, 17, 1413 7 of 17

The calibration set is shown in green and the validation set in red. The calibration process uses
the calibration set to adjust the distribution of the daily SPP data to match the distribution of the daily
rain gauge data [62]. A function is set to adjust the precipitation data from a rainfall threshold set at
1 mm/d. The cut-off threshold is used to remove low precipitation values in the SPP data in order to
equalize the frequency of wet days between the daily rain gauge and SPP precipitation data sets [63].
The result obtained corresponds to correction coefficients, calculated, for each percentile. The greater
the number of quantile divisions used to represent the underlying frequency distributions, the better
the correction [64]. The correction coefficients are calculated from the daily mean rain gauge series
and the daily mean satellite series of the calibration set. In a second step, the correction coefficients
obtained from the calibration set are applied to correct the SPP series of the validation set (red pixel in
Figure 4). We implemented the same leave-one-out cross validation used in the paper of Kim et al. [63]:
correction coefficients are calculated over 11 years and then applied to correct one year (the omitted
year) of the SPP series of the validation set. This procedure is repeated for every year. The 12 years
obtained by the validation process are combined into a single time series that forms the corrected
SPP series.
In the example shown in Figure 4, the correction coefficients obtained from the calibration set
(green pixels) are applied to the SPP series of the validation set (red pixel) to obtain the corrected SSP
series. In each simulation on the 93 daily SPP series representing the available rain gauges, 70 daily
SPP series are used as the calibration set, and 23 are selected as the validation set for the QM correction.
The 23 daily SPP series of the validation set are chosen by selecting one rain gauge in each of the
23 hydroclimatic areas and the rest are used as calibration set. Comparison of the relative bias (rBIAS)
and the relative RMSE (rRMSE) between the different simulations of the spatial clustering is performed
on these same validation set.

4. Results
For each of the three simulations undertaken, we compared both the TRMM-TMPA 3B42V7
precipitation estimates against the daily rain gauge measurements, and the TRMM-TMPA 3B42V7
precipitation estimates corrected with the QM method against the daily rain gauge measurements.
The accuracy of the QM correction method in the various configurations presented above was
evaluated via the calculation of the deviations: estimated – measured, and using statistical indices:
rBIAS and rRMSE (Table 1).

Table 1. Quantitative statistical criteria. For a given rain gauge and its associated pixel, Oi is the rain
gauge value and Ei the satellite value. N is the total number of days in the time series.

Statistical Criteria Formula

N
1
BIAS N ∑ ( Ei − Oi )
i =1
Biais
rBIAS O
s
N
1 2
RMSE N ∑ ( Ei − Oi )
i =1
RMSE
rRMSE O

4.1. Quality of Corrected TRMM-TMPA 3B42V7 Estimates for the Entire Study Area as a Calibration Set
In the first simulation, the parameterization of the QM correction method is performed for the
entire zone using data from the 70 rain gauges in the calibration set.
Sensors 2017, 17, 1413 8 of 17

4.1.1. Global Assessment

Figure 5 shows the global rBIAS and global rRMSE for the 23 validation pixels considered as a
whole. The biases and RMSEs are obtained after comparison of both the precipitation estimated from
TRMM-TMPA 3B42V7 and daily rain gauge measurements (black), and the precipitation corrected
with
Sensorsthe QM
2017, 17, method
1413 and daily rain gauge measurements (grey). 8 of 17

Sensors 2017, 17, 1413 8 of 17

Figure 5. (a) Global relative bias (rBIAS; %) and (b) global relative RMSE (rRMSE; %) obtained by
comparing the
comparing the precipitation
precipitation estimated
estimated from TRMM-TMPA
from TRMM-TMPA 3B42V7 with3B42V7 withgauge
daily rain dailymeasurements
rain gauge
Figure
measurements 5. (a) Global
(black), relative
and thebias (rBIAS; %)
precipitation and (b) global
corrected withrelative
the QMRMSE (rRMSE;
method with%)
(black), and the precipitation corrected with the QM method with daily rain gauge measurements obtained
daily rainby
gauge
comparing
measurements the precipitation
(grey) study
over the estimated from
entire study area. TRMM-TMPA 3B42V7 with daily rain gauge
(grey) over the entire area.
measurements (black), and the precipitation corrected with the QM method with daily rain gauge
measurements (grey) over the entire study area.
Figure 5a
Figure 5a shows
shows that
thatthetheglobal
globalrBIAS
rBIASisisworse
worse after thethe
after QM QM correction.
correction. Without
Withoutcorrection, the
correction,
average of the
Figure
the average rBIAS of
5a shows
of the the
rBIASthat 23
of the daily
the globalSPP
23 daily series
rBIASSPP is
is worse −3% against
is −the
seriesafter 3%QM −25% after the
−25%Without
correction.
against QM correction.
after thecorrection, For
the
QM correction.the
For the rRMSE, Figure 5b shows a very slight improvement of only 1% for the QM-correctedresult
rRMSE, Figure
average of 5b
the shows
rBIAS a
of very
the slight
23 daily improvement
SPP series is of
−3% only 1%
against for
−25% the QM-corrected
after the QM data.
correction. This
For the data.
shows
This thatshows
rRMSE,
result for the
Figure 5bGuiana
thatshows Shield,
a very
for the the
slight
Guiana TRMM-TMPA
improvement
Shield, of only3B42V7
the TRMM-TMPA 1% for theSPP provides
QM-corrected
3B42V7 adata.
SPP provides reasonably
This resultgood
a reasonably
estimation
good shows ofthatprecipitation
estimation for the Guiana on Shield,
of precipitation the global
on the scale.
the TRMM-TMPA
global scale.TheTherBIAS
3B42V7
rBIAS degradation
SPP provides
degradation ofofthe QM-corrected
a reasonably
the QM-correctedgood data
data
estimation
indicates that of
this precipitation
method is on
not the global
applicable scale.
to large The rBIAS
spatial degradation
scales. of the QM-corrected data
indicates that this method is not applicable to large spatial scales.
indicates that this method is not applicable to large spatial scales.
4.1.2. Local
4.1.2. Local Assessment
Assessment
4.1.2. Local Assessment
Figure 66 shows
Figure shows thethe rBIASs
rBIASs and rRMSEs
rRMSEs for thethe 2323 validation
validation daily
daily SPP
SPP series
series considered
considered
Figure 6 shows the rBIASsand and rRMSEs forfor the 23 validation daily SPP series considered
individually.
individually. The
individually.
rBIASs
The The
rBIASs and
and
rBIASs and
rRMSEs
rRMSEs
rRMSEsare
are obtained
areobtained
after
obtained after
comparing
comparingthe
after comparing
the
the precipitation
precipitation
precipitation
estimated
estimated
estimated
from
fromfrom
TRMM-TMPA
TRMM-TMPA
TRMM-TMPA 3B42V7
3B42V7
3B42V7andand
and daily
daily rain
rain
daily gauge
raingauge measurements
gaugemeasurements (black),and
(black),
measurements (black), and
and the
thethe precipitation
precipitation
precipitation corrected
corrected
corrected
with the
with the QM
withQM method
method
the QM and
andand
method daily
daily rain
rain
daily gauge
gauge
rain measurements
gaugemeasurements (grey).
(grey).
measurements (grey).

Figure
Figure 6. Relative
6. Relative biasbias (rBIAS;
(rBIAS; %)%) andrelative
and relativeRMSE
RMSE (rRMSE;
(rRMSE;%) %)obtained
obtainedbyby
comparison of both
comparison the the
of both
precipitation estimated from TRMM-TMPA 3B42V7 and daily rain gauge measurements (black), and
precipitation estimated from TRMM-TMPA 3B42V7 and daily rain gauge measurements (black), and
Figure
the6.precipitation
Relative bias (rBIAS;with
corrected %) and relative
the QM RMSE
method and(rRMSE;
daily rain%) obtained
gauge by comparison
measurements of both
(grey) for the the
the precipitation corrected with the QM method and daily rain gauge measurements (grey) for the
precipitation
validationestimated from
pixel in each TRMM-TMPA
of the 3B42V7
23 hydroclimatic areas.and daily rain gauge measurements (black), and
validation pixel in each of the 23 hydroclimatic areas.
the precipitation corrected with the QM method and daily rain gauge measurements (grey) for the
Figurepixel
validation 6 shows
in eachthat only
of the 23 four daily SPPareas.
hydroclimatic series have a low bias without applying the QM
correction. When the QM correction is applied throughout the entire study area, seven of the 23
validation daily SPP series show a correction in rBIAS and 16 daily SPP series show a deterioration
Figure 6 shows that only four daily SPP series have a low bias without applying the QM
in rBIAS in comparison with the raw data. For the seven daily SPP series corrected, the improvement
correction. When the QM correction is applied throughout the entire study area, seven of the 23
of rBIAS is on average 13.4%. Conversely, for the 16 uncorrected daily SPP series, the deterioration
validation daily SPP series show a correction in rBIAS and 16 daily SPP series show a deterioration
Sensors 2017, 17, 1413 9 of 17

Figure 6 shows that only four daily SPP series have a low bias without applying the QM correction.
When the QM correction is applied throughout the entire study area, seven of the 23 validation daily
SPP series show a correction in rBIAS and 16 daily SPP series show a deterioration in rBIAS in
comparison with the raw data. For the seven daily SPP series corrected, the improvement of rBIAS is
on average
Sensors 13.4%.
2017, 17, 1413 Conversely, for the 16 uncorrected daily SPP series, the deterioration of rBIAS is
9 of 17
on average 16.7%. Regarding the rRMSE, after the QM correction, the 23 daily SPP series show lower
show lower
rRMSEs withrRMSEs withimprovement
an average an average improvement
of 40%. These of 40%.show
results Thesethat
results show
for the that Shield,
Guiana for the the
Guiana
QM
correction increases the bias for a large number of daily SPP series and thus, it is not a method thatnot
Shield, the QM correction increases the bias for a large number of daily SPP series and thus, it is is
a method that
applicable is applicable
to large to large spatial scales.
spatial scales.

4.2. Quality of Corrected TRMM-TMPA

TRMM-TMPA 3B42V7
3B42V7 Estimates
Estimates for
for 66 Hydroclimatic
Hydroclimatic Areas
Areasas
asCalibration
CalibrationSets
Sets
In the second simulation, we apply the HAC method
method to all the rain gauge data to obtain 6 classes
of hydroclimatic regime. The QM correction method is then parameterized with data contained in
each hydroclimatic
hydroclimaticregime.
regime.The
Theparameterization
parameterizationofofthe QM
the QM correction method
correction methodis done for for
is done eacheach
of the
of
6the
hydroclimatic zones,
6 hydroclimatic and these
zones, parameters
and these are applied
parameters to thetovalidation
are applied pixelspixels
the validation located in eachinofeach
located the
hydroclimatic zones. zones.
of the hydroclimatic
Figure 7 shows the rBIASs
rBIASs and
and rRMSEs
rRMSEs for
for the
the validation
validation daily
daily SPP
SPP series.
series. The rBIASs and
rRMSEs are obtained by comparing both the TRMM-TMPA 3B42V7 precipitation estimates and daily
rain gauge measurements, and the TRMM-TMPA 3B42V7 precipitation estimates corrected with the
QM method and daily rain gauge measurements.
measurements.

Figure 7. (a)
(a) Relative
Relativebias
bias(rBIAS;
(rBIAS;%)
%)and
and(b)
(b)relative
relativeRMSE
RMSE(rRMSE;
(rRMSE; %)%)obtained byby
obtained comparison
comparisonof
both
of thethe
both precipitation
precipitationestimated
estimatedfrom
fromTRMM-TMPA
TRMM-TMPA3B42V7 3B42V7and
anddaily
daily rain
rain gauge measurements
(black), and
(black), andthe
theprecipitation
precipitation corrected
corrected with
with thethe
QMQM method
method and and
dailydaily rain gauge
rain gauge measurements
measurements (grey)
(grey)
for for 23 in
23 pixels pixels in 6 hydroclimatic
6 hydroclimatic areas. areas.

In comparison
In comparison with with the
the raw
raw data,
data, nine
nine of
of the
the 23
23 validation
validation daily
daily SPP
SPP series
series show
show aa correction
correction in
in
rBIAS and 14 daily SPP series show a deterioration in rBIAS. The improvement of
rBIAS and 14 daily SPP series show a deterioration in rBIAS. The improvement of rBIAS for the ninerBIAS for the nine
validation daily
validation daily SPP
SPP series
series is
is on
on average
average 14.2%.
14.2%. The
The deterioration of rBIAS
deterioration of rBIAS for
for the
the other
other 1414 validation
validation
daily SPP series is on average 9.7%. Compared with the first simulation, it is evident
daily SPP series is on average 9.7%. Compared with the first simulation, it is evident that only two that onlymore
two
more daily SPP series show an improvement after QM. The improvement of the
daily SPP series show an improvement after QM. The improvement of the nine validation daily SPPnine validation daily
SPP series
series is slightly
is slightly better
better thanthan inprevious
in the the previous simulation.
simulation. However,
However, the degradation
the degradation of theof14
the 14 daily
daily SPP
SPP series after the QM correction is much less significant when taking into account the six
hydroclimatic zones rather than considering the entire region as a single hydroclimatic area.
Regarding the rRMSEs, after QM correction, 23 daily SPP series show lower rRMSEs with an
average improvement of 40%. The improvement is of the same order of magnitude as the first
simulation. These results show that for the Guiana Shield, the correction by QM shows a bias
degradation for a large number of daily SPP series and that it is therefore not applicable to large
Sensors 2017, 17, 1413 10 of 17

series after the QM correction is much less significant when taking into account the six hydroclimatic
zones rather than considering the entire region as a single hydroclimatic area.
Regarding the rRMSEs, after QM correction, 23 daily SPP series show lower rRMSEs with an
average improvement of 40%. The improvement is of the same order of magnitude as the first
simulation. These results show that for the Guiana Shield, the correction by QM shows a bias
degradation for a large number of daily SPP series and that it is therefore not applicable to large
spatial scales.
SensorsThese
2017, 17,results
1413 show that dividing the study area into six different hydroclimatic areas and
10 of 17
correcting the bias using the parameterized QM correction method with the data contained in each
of
of the
the hydroclimatic
hydroclimaticzones
zonesmakes
makesititpossible
possibletotoimprove
improvethetherBIAS
rBIASof ofthe
theTRMM-TMPA
TRMM-TMPA 3B42V7 3B42V7 SPPSPP
slightly. However,
slightly. However, more than half the
the daily SPP series remain uncorrected. On the regional scale, the
daily SPP series remain uncorrected. On the regional scale, the
TRMM-TMPA
TRMM-TMPA 3B42V7 3B42V7 SPP
SPPshows
showsgreater
greaterdifficulty
difficultyinin correctly
correctly estimating
estimating precipitation.
precipitation. Although
Although the
the
QMQM correction
correction makesmakes it possible
it possible to some
to correct correct some this
regions, regions, this method
correction correction
stillmethod still has
has difficulties in
difficulties in improving the bias
improving the bias in certain regions. in certain regions.

4.3. Quality
4.3. Qualityofofcorrected
corrected TRMM-TMPA
TRMM-TMPA 3B42V7
3B42V7 Estimates
Estimates for
for 23 Hydroclimatic Areas as Calibration Sets
In this
In this third
third simulation,
simulation, the the QM
QM correction
correction method
method isis applied
applied to
to the
the 23
23 hydroclimatic
hydroclimatic areas
areas
obtained by
obtained by the
the HAC
HAC method
method and and represented
represented inin Figure
Figure 8.
8. For
Foreach
eachof
ofthe
the23
23hydroclimatic
hydroclimaticareas,
areas,
one pixels
one pixels is
is used
used as
as the
the validation
validation daily
daily SPP
SPP series.
series. The
Theremaining
remainingpixels
pixelsinineach
eacharea
area are
are used
used as
as
calibration daily
calibration daily SPP
SPP series
series for
for the
the QM
QM correction
correction method.
method.

Figure 8. Spatial
Figure 8. Spatialmap
map of 23
of the thehydroclimatic
23 hydroclimatic areas obtained
areas obtained by the hierarchical
by the hierarchical ascendant
ascendant classification
classification method. Black dots represent the validation
method. Black dots represent the validation pixels. pixels.

The parameterization of the QM correction method is done for each of the 23 hydroclimatic
The parameterization of the QM correction method is done for each of the 23 hydroclimatic areas,
areas, and these parameters are applied to the validation daily SPP series located in each
and these parameters are applied to the validation daily SPP series located in each hydroclimatic area.
hydroclimatic area. The rBIASs and rRMSEs are calculated by comparing the satellite data with the
The rBIASs and rRMSEs are calculated by comparing the satellite data with the QM corrected data
QM corrected data (Figure 9). Figure 9 shows the results of the rBIASs and rRMSEs for each of the 23
(Figure 9). Figure 9 shows the results of the rBIASs and rRMSEs for each of the 23 validation daily SPP
validation daily SPP series. Compared with the raw satellite data, 18 of the 23 validation daily SPP
series. Compared with the raw satellite data, 18 of the 23 validation daily SPP series show a correction
series show a correction of rBIAS and only five daily SPP series show a worsening of rBIAS. The
of rBIAS and only five daily SPP series show a worsening of rBIAS. The improvement of rBIAS for the
improvement of rBIAS for the corrected daily SPP series is on average 12.1%. The deterioration of
corrected daily SPP series is on average 12.1%. The deterioration of rBIAS for the degraded daily SPP
rBIAS for the degraded daily SPP series is on average 9.8%. Compared with the first two simulations,
series is on average 9.8%. Compared with the first two simulations, the improvement in rBIAS is much
the improvement in rBIAS is much better, with 78% of validation daily SPP series showing a
better, with 78% of validation daily SPP series showing a correction of rBIAS after QM correction.
correction of rBIAS after QM correction.
Sensors 2017, 17, 1413 11 of 17
Sensors 2017, 17, 1413 11 of 17

Figure 9.
Figure 9. (a)
(a) Relative
Relative bias
bias (rBIAS;
(rBIAS;%)%)and
and(b)
(b)relative
relativeRMSE
RMSE(rRMSE;
(rRMSE;%)%)obtained byby
obtained comparison
comparisonof
both the precipitation estimated from TRMM-TMPA 3B42V7
of both the precipitation estimated from TRMM-TMPA 3B42V7 and daily and daily rain gauge measurements
gauge measurements
(black),and
(black), andthetheprecipitation
precipitation corrected
corrected with
with the the
QMQM method
method and daily
and daily rain gauge
rain gauge measurements
measurements (grey)
(grey)
for for 23using
23 pixels pixels23
using 23 hydroclimatic
hydroclimatic areas. areas.

Regarding the rRMSEs, after QM correction, 20 daily SPP series show lower rRMSEs with an
Regarding the rRMSEs, after QM correction, 20 daily SPP series show lower rRMSEs with an
average improvement of 28%. Only three daily SPP series show larger rRMSEs after QM correction.
average improvement of 28%. Only three daily SPP series show larger rRMSEs after QM correction.

4.4. Performance
4.4. Performance
Figure 10
Figure 10 shows
shows thethe Empirical
Empirical Cumulative
Cumulative Distribution
Distribution Function
Function (ECDF)
(ECDF) of of daily
daily rain
rain gauge
gauge
data, SPP data and of corrected SPP data for the three simulations under
data, SPP data and of corrected SPP data for the three simulations under consideration. In the first consideration. In the first
simulationthe
simulation theparameterization
parameterizationofofthe theQM QM correction
correction method
method is done
is done for entire
for the the entire
zonezonewith with the
the data
data from the 70 rain gauges in the calibration set. The ECDF of
from the 70 rain gauges in the calibration set. The ECDF of the corrected SPP shows no improvementthe corrected SPP shows no
improvement and seems to be moving away from the ECDF of the daily rain
and seems to be moving away from the ECDF of the daily rain gauge data, in particular for intensities gauge data, in particular
for intensities
greater than 15greater
mm/d.than The15 mm/d.bias
relative Theofrelative bias of SPP
the corrected the corrected
(−31.74%)SPP (−31.74%)
is about is aboutthan
1% greater 1%
greater than the relative bias before QM correction (−30.61%). For intensities
the relative bias before QM correction (−30.61%). For intensities less than 15 mm/d, i.e., 80% of the less than 15 mm/d, i.e.,
80% of the
recorded recorded precipitation,
precipitation, the correctionthe correction effective.
is remarkably is remarkablyHowever,effective. However,between
for intensities for intensities
15 and
between 15 and 40 mm/d, the ECDF of the corrected SPP data is largely
40 mm/d, the ECDF of the corrected SPP data is largely greater than the gauge’s ECDF. High intensities greater than the gauge’s
ECDF.
are High intensities are not corrected.
not corrected.
In the
In thesecond
secondsimulation,
simulation,the theQMQMcorrection
correction method
method is parameterized
is parameterized with
with datadata
from from
eacheach
of theof
the six hydroclimatic regimes. The relative bias of the SPP data improved
six hydroclimatic regimes. The relative bias of the SPP data improved markedly from −30.61% before markedly from −30.61%
before
QM QM correction
correction to −13.01%to −13.01%
after QM after QM correction.
correction. For intensities
For intensities less thanless20 than
mm/d, 20 i.e.,
mm/d, i.e.,85%
about aboutof
85% of precipitation,
precipitation, the QM correction
the QM correction is effective.isFor
effective.
intensitiesFor between
intensities between
20 and 20 and
50 mm/d, the50ECDF
mm/d, the
of the
ECDF
SPP of correction
after the SPP after correction
is still is still afrom
a little removed littlethe
removed
ECDF offrom the ECDF
the daily of themeasurement.
rain gauge daily rain gauge
measurement.
In the third simulation, the QM correction method is parameterized with data from each of the
In the third regimes.
23 hydroclimatic simulation, Thethe
ECDF QM of correction method
the corrected SPPisdata
parameterized
shows betterwith data from
agreement with each
the of the
daily
23 hydroclimatic regimes. The ECDF of the corrected SPP data shows
rain gauge compared to the ECDF of SPP data before QM correction. Indeed, the relative bias is better agreement with the daily
rain gaugefrom
improved compared
−30.61%totothe ECDF For
−9.83%. of SPP data before
intensities less thanQM 25correction.
mm/d, i.e.,Indeed,
about 90% the of
relative bias is
precipitation,
improved from −30.61% to −9.83%. For intensities less than 25 mm/d, i.e.,
the QM correction shows very good results. However, the tendency to correct the intensity between 25 about 90% of precipitation,
the QM
and correction
50 mm/d is moreshows very good results. However, the tendency to correct the intensity between
difficult.
25 and 50 mm/d is more difficult.
Sensors 2017, 17, 1413 12 of 17
Sensors 2017, 17, 1413 12 of 17

Figure 10. Empirical

Figure10. Empirical Cumulative
CumulativeDistribution
DistributionFunction
Functionfor
foraarain
raingauge
gaugeandandits
itsassociated
associatedpixel
pixelunder
under
the
thethree
threesimulations.
simulations. Each
Each simulation
simulation shows
shows the
the ECDF
ECDF ofof the
the daily
daily rain
raingauge
gauge(black),
(black), the
theECDF
ECDFof
of
SPP
SPPbefore
beforeQM
QM(red)
(red)and
andthe
theECDF
ECDFofofSPP
SPPafter
afterQM
QM(blue).
(blue).

In summary,
In summary, the
the QM
QM method
method corrects
corrects the
the weak
weak and
and high
high intensities
intensities rather
rather than
than the
the medium
medium
intensities. The results show this novel two set approach reduces bias and RMSE significantly.
intensities. The results show this novel two set approach reduces bias and RMSE significantly.

5. Discussion
5. Discussion and
and Conclusions
Conclusions
The objective was to
The to demonstrate
demonstratethe theimportance
importanceofof considering
considering hydroclimatic
hydroclimatic regimes
regimesin the
in
correction
the of daily
correction satellite
of daily precipitation
satellite data. We
precipitation used
data. Wea spatial
used a classification approachapproach
spatial classification and exploited
and
the QM method
exploited the QM tomethod
correct daily TRMM-TMPA
to correct 3B42V7 SPP3B42V7
daily TRMM-TMPA estimates. SPPThe rain gauge
estimates. The measurement
rain gauge
pixels were grouped
measurement spatially
pixels were into different
grouped spatiallyhydroclimatic
into differentzones according zones
hydroclimatic to theiraccording
temporal to profiles
their
based onprofiles
temporal long-term basedmonthly
on long-term means of precipitation
monthly data usingdata
means of precipitation theusing
HACthemethod. HAC method. Then,
parameterization
Then, parameterization of the of QM themethod
QM method was performed
was performed for each for of
eachtheofhydroclimatic
the hydroclimatic areas,areas,
and these
and
parameters were used to correct the daily TRMM-TMPA 3B42V7
these parameters were used to correct the daily TRMM-TMPA 3B42V7 SPP estimates. In order show SPP estimates. In order to show
howvariation
how variationin inthe
the scale
scale ofof analysis
analysis affected
affected thethe rBIAS
rBIAS and
and rRMSE,
rRMSE, threethree simulations
simulations with with different
different
spatial divisions
spatial divisions were wereconducted,
conducted, and and thethe effects
effectsof ofthe
theQMQMmethodmethod on onthe
theaccuracy
accuracy of ofthe
thecorrected
corrected
SPPestimates
SPP estimateswerewereobserved.
observed.
Forthe
For theglobal
global scale
scale (simulation
(simulation 1), 1), applying
applying the the QM
QM correction
correction TRMM-TMPA
TRMM-TMPA3B42V7 3B42V7has hasaapoor
poor
effect: only seven out of 23 daily SPP series are corrected. The approach
effect: only seven out of 23 daily SPP series are corrected. The approach used in this paper, which used in this paper, which
appliesthe
applies thecorrection
correctionto to the
the satellite
satellite series
series from
from the
the calibration
calibration dailydaily SPP
SPP series
series throughout
throughout the the zone,
zone,
doesnot
does notmake
makeit itpossible
possible to to correct
correct thethebiasbias present
present in SPP.
in the the SPP.
The QM Thecorrection
QM correction method, method, on a
on a global
global scale, is limited in its ability to correct atmospheric phenomena
scale, is limited in its ability to correct atmospheric phenomena that occur at finer scale. Even after that occur at finer scale. Even
after reducing
reducing the spatial
the spatial scale to scale to 6 hydroclimatic
6 hydroclimatic groupsgroups
(simulation(simulation 2), the
2), the QM QM correction
correction appliedapplied
within
within
each each hydroclimatic
hydroclimatic area stillarea
showsstill significant
shows significant
relativerelative
biases. biases.
Only nine Only outnine
of 23 outdaily
of 23SPPdaily SPP
series
series have an improved relative bias after the QM correction. However,
have an improved relative bias after the QM correction. However, the daily SPP serie that are not the daily SPP serie that are
not corrected
corrected havehave
smaller smaller
errorserrors
than inthan in simulation
simulation 1. In order
1. In order to improve
to improve the relative
the relative bias, we bias, weagain
once once
reduced the spatial scale in order to correct daily SPP series within a smaller area that is characteristicis
again reduced the spatial scale in order to correct daily SPP series within a smaller area that
characteristic
of a more precise of aclimate
more precise
regime.climate
Simulation regime. Simulation
3 divides the area 3 divides
into 23 the area into 23groups.
hydroclimatic hydroclimatic
At this
groups.
finer Atbias
scale, thisresults
finer scale, bias resultsimproved.
are significantly are significantly
Of the improved.
23 daily SPP Ofseries,
the 23which
daily correspond
SPP series, whichto the
correspond to the 23 areas, 18 show an improvement in rBIAS,
23 areas, 18 show an improvement in rBIAS, and 20 show an improvement in rRMSE. Applying the and 20 show an improvement in
rRMSE.
QM Applying
correction methodthe QMto thecorrection methodintoeach
data contained the data containedarea
hydroclimatic in each
makes hydroclimatic
it possible to area makes
improve
it possible
the accuracy toofimprove the accuracy
the TRMM-TMPA of the TRMM-TMPA
3B42V7 SPP at the finer3B42V7 SPP at the finer spatial scale.
spatial scale.
The principal
The principal finding
finding was was that
that thethe QM
QM biasbias correction
correction performed
performed worst worst at at the
the coarsest
coarsest scale,
scale,
whereas additional
whereas additional benefits
benefits werewere produced
produced at at aa finer
finerscale.
scale. Classifying
Classifying the the rain
rain gauge
gauge datadata byby
hydroclimatic area clearly improved the rBIASs and rRMSEs,
hydroclimatic area clearly improved the rBIASs and rRMSEs, reducing errors in the precipitationreducing errors in the precipitation
estimates. The
estimates. The temporal
temporal characteristics
characteristics of of the
the sampling
sampling units,
units, such
such as as amplitude
amplitude or or similarity
similarity of of the
the
time series, are important scale concepts. The analysis of the ECDF shows that the QM method
corrects the weak intensities rather than the medium intensities.
Sensors 2017, 17, 1413 13 of 17

time series, are important scale concepts. The analysis of the ECDF shows that the QM method corrects
the weak intensities rather than the medium intensities.
It is important to note a deterioration of rBIAS after QM correction for certain SPP series. The daily
SPP series with a degradation of rBIAS can be classified according to two cases. The first case are SPP
series that considerably overestimate or underestimate precipitation before QM correction, which is
amplified after QM correction. The second case is SPP series that have low rBIAS before QM correction,
which see their rBIAS degraded after QM correction. This category of SPP series does not require
QM correction because the satellite series is very close to the gauge series. For example, within the
23 hydroclimatic areas, only five daily SPP series showed degradation. Of these five, three showed
rBIASs before correction of <1%. Therefore, only two daily SPP series showed a large degradation
of rBIAS.
A multitude of studies show that the performance of SPPs is a function of precipitation intensities,
climate and seasons. Simulation 1 calibrates the correction method from 70 rain gauges available
throughout the region. Calibration is therefore carried out on pixels which may have a precipitation
difference of more than 2000 mm/year. Even after subdividing the area into 6 regions, (Simulation 2),
the correction’s effectiveness is evidently still limited. The more we decompose into hydroclimatic
zones, the more we group together, within a zone, rain gauges that are very close climatically.
This grouping improves the efficiency of the QM correction to reduce the bias. Consequently, if
the areas that are used to define the correction parameters contain more homogeneous time data, the
more likely it is that the quality of the correction is high. This is because the correction will apply to a
SPP series with a time profile “similar” to those used for the calibration of the correction parameters.
Conversely, if the correction parameters are established from areas containing more heterogeneous
data, it is more likely that the correction parameters will be poorly adapted to the SPP series to be
corrected, and hence the quality of the correction will be of low quality. This shows the importance of
taking into account spatial scale in the QM correction method.
Several hypotheses may be invoked to explain this degradation. A first hypothesis is that the
method compares and corrects a satellite pixel covering 625 km2 with a rain gauge point. There is
therefore an incompressible bias between the amount of precipitation at the rain gauge point and the
amount of precipitation in the satellite pixel. This is linked to localized phenomena such as convective
rain, which can be very intense and spatially restricted. A second hypothesis is related to the criterion
used to define the hydroclimatic zones, i.e., the 12 monthly averages of precipitation data. This criterion
can lead to bias degradation linked to different temporal phenomena existing during the 12 years of
the time series, but it can also lead to identical hydroclimatic regimes when monthly averages are used.
From a spatial perspective, studies have been carried out on the capacity of SPP to estimate
precipitation in mountainous areas. Zambrano-Bigiarini et al. [65] compare 14 regions across Chile
and show that SPPs perform better in low and mid-altitude regions. In our study, the altitude of the
23 daily rain gauges associated to the 23 validation daily SPP series varies from 1 m (Chaves and
Kourou) to 305 m (Lourenco). We observe, with the third simulation, that among the five lowest daily
rain gauges and associated SPP series (>10 m), only one sees its bias increase after QM correction.
Conversely, the five highest daily rain gauges and associated SPP series (>100 m), all show a decrease
in bias after QM correction in the associated daily SPP series. The five daily SPP series that see their
bias deteriorate after QM correction, in simulation 3, are at low and mid-altitudes of 1, 12, 16, 71 and
76 m. These results show that the topography of our region does not explain the worsening of the
bias. The second possible explanation of the bias differences after correction is distance to the coast.
Among the five daily SPP series that show a deterioration of the bias after correction in the third
simulation, the distances to the coast are very variable. The validation daily rain gauge in the area
5c2 (Fazenda Parana) is 640 km from the coast, while the validation daily rain gauge in the areas
5b11 (Saint Laurent) and 4a1 (Chaves) are located at a distance of just 20 km and 2 km, respectively.
The distance to the coast cannot therefore be used as a criterion to justify the degradation at these
daily SPP series. In terms of land cover, of our 23 validation daily rain gauge associated to daily SPP
Sensors 2017, 17, 1413 14 of 17

series, 7 are at a distance of less than 1 km from the lake or river, of which four show a deterioration
of the bias after correction, i.e., more than half. In the microwave domain, the radiometer receives
the signal of the diffusion and absorption of hydrometeors but also of the terrestrial surfaces, thus
measuring the moisture content of these surfaces. Our results may therefore show that the proximity of
water has an influence on the difficulty of the SPP in the estimation of the precipitations and thus the
difficulty of correctly correcting certain daily SPP series. The radiometer measures moisture content
above water bodies and thus overestimates precipitation. These results are confirmed by Delahaye [8]
which observes errors of estimation over dense forests, due to the high moisture content captured by
the radiometer.
In addition to a spatial approach, a temporal approach could be considered. In other words, the
satellite data could be corrected according to season or according to other criteria such as the temporal
characteristics of rainfall [63]. This would make it possible to divide the time series into several time
sequences, and to correct each of the sequences according to different correction coefficients.
Our study has indicated that hydroclimatic classification is relevant for establishing a bias
correction method. We conclude that this novel two set approach to correct for TRMM-TMPA 3B42V7
SPP estimates provides acceptable results for the Guiana Shield. The perspective ultimately aims to
use these results to design a methodology for correcting SSP pixels on a wider scale with a limited
amount of gauge data. Future work will investigate the possibility of correcting TRMM-TMPA 3B42V7
SPPs based on predefined hydroclimatic areas. Thus, each satellite pixel will be associated with a
hydroclimatic regime and will be corrected based on the rain gauge data present within the same
hydroclimatic regime. This work will make it possible to improve the correction of TRMM-TMPA
3B42V7 SPPs at the large scale for areas with a sparse rain gauge measurement network.

Acknowledgments: The research leading to these results has received funding from Fonds Européen de
Développement Régional (FEDER-Guyane/PHEUS Project). We acknowledge Météo France and Agência Nacional
de Àguas for providing the daily precipitation data. Finally, special thanks to the original producers of all evaluated
precipitation products for providing free downloadable precipitation products. We would like to thank Editage
(www.editage.com) for English language editing.
Author Contributions: Justine Ringard, Laurent Linguet and Frederique Seyler conceived and designed the
experiments; Justine Ringard performed the experiments; Justine Ringard analyzed the data and wrote the paper.
All authors commented the manuscript and contributed to the discussion and conclusions.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Tapiador, F.J.; Turk, F.J.; Petersen, W.; Hou, A.Y.; García-Ortega, E.; Machado, L.A.T.; Angelis, C.F.; Salio, P.;
Kidd, C.; Huffman, G.J.; et al. Global precipitation measurement: Methods, datasets and applications.
Atmos. Res. 2012, 104–105, 70–97. [CrossRef]
2. Derin, Y.; Yilmaz, K.K. Evaluation of Multiple Satellite-Based Precipitation Products over Complex
Topography. J. Hydrometeorol. 2014, 15, 1498–1516. [CrossRef]
3. Huffman, G.J.; Bolvin, D.T.; Nelkin, E.J.; Wolff, D.B.; Adler, R.F.; Gu, G.; Hong, Y.; Bowman, K.P.; Stocker, E.F.
The TRMM Multisatellite Precipitation Analysis (TMPA): Quasi-Global, Multiyear, Combined-Sensor
Precipitation Estimates at Fine Scales. J. Hydrometeorol. 2007, 8, 38–55. [CrossRef]
4. Huffman, G.J.; Adler, R.F.; Bolvin, D.T.; Nelkin, E.J. The TRMM Multi-Satellite Precipitation Analysis (TMPA).
In Satellite Rainfall Applications for Surface Hydrology; Gebremichael, M., Hossain, F., Eds.; Springer: Dordrecht,
The Netherlands, 2010; pp. 3–22.
5. Huffman, G.J.; Bolvin, D.T. TRMM and Other Data Precipitation Data Set Documentation.
Available online: https://pmm.nasa.gov/sites/default/files/document_files/3B42_3B43_doc_V7.pdf
(accessed on 8 April 2015).
6. Joyce, R.J.; Janowiak, J.E.; Arkin, P.A.; Xie, P. CMORPH: A Method that Produces Global Precipitation
Estimates from Passive Microwave and Infrared Data at High Spatial and Temporal Resolution.
J. Hydrometeorol. 2004, 5, 487–503. [CrossRef]
Sensors 2017, 17, 1413 15 of 17

7. Sorooshian, S.; Hsu, K.-L.; Gao, X.; Gupta, H.V.; Imam, B.; Braithwaite, D. Evaluation of PERSIANN System
Satellite–Based Estimates of Tropical Rainfall. Bull. Am. Meteorol. Soc. 2000, 81, 2035–2046. [CrossRef]
8. Delahaye, F. Analyse Comparative des Différents Produits Satellitaires D’estimation des Précipitations en
Amazonie Brésilienne. Ph.D. Thesis, University of Rennes II, Rennes, France, 2013.
9. Guo, H.; Chen, S.; Bao, A.; Hu, J.; Gebregiorgis, A.; Xue, X.; Zhang, X. Inter-Comparison of High-Resolution
Satellite Precipitation Products over Central Asia. Remote Sens. 2015, 7, 7181–7212. [CrossRef]
10. Pereira Filho, A.J.; Carbone, R.E.; Janowiak, J.E.; Arkin, P.; Joyce, R.; Hallak, R.; Ramos, C.G.M. Satellite
Rainfall Estimates Over South America—Possible Applicability to the Water Management of Large
Watersheds. J. Am. Water Resour. Assoc. 2010, 46, 344–360. [CrossRef]
11. Ringard, J.; Becker, M.; Seyler, F.; Linguet, L. Temporal and Spatial Assessment of Four Satellite Rainfall
Estimates over French Guiana and North Brazil. Remote Sens. 2015, 7, 16441–16459. [CrossRef]
12. Tan, M.; Ibrahim, A.; Duan, Z.; Cracknell, A.; Chaplot, V. Evaluation of Six High-Resolution Satellite and
Ground-Based Precipitation Products over Malaysia. Remote Sens. 2015, 7, 1504–1528. [CrossRef]
13. Thiemig, V.; Rojas, R.; Zambrano-Bigiarini, M.; Levizzani, V.; De Roo, A. Validation of Satellite-Based
Precipitation Products over Sparsely Gauged African River Basins. J. Hydrometeorol. 2012, 13, 1760–1783.
[CrossRef]
14. Toté, C.; Patricio, D.; Boogaard, H.; van der Wijngaart, R.; Tarnavsky, E.; Funk, C. Evaluation of Satellite
Rainfall Estimates for Drought and Flood Monitoring in Mozambique. Remote Sens. 2015, 7, 1758–1776.
[CrossRef]
15. Vila, D.A.; de Goncalves, L.G.G.; Toll, D.L.; Rozante, J.R. Statistical Evaluation of Combined Daily Gauge
Observations and Rainfall Satellite Estimates over Continental South America. J. Hydrometeorol. 2009, 10,
533–543. [CrossRef]
16. Kidd, C. Satellite rainfall climatology: A review. Int. J. Climatol. 2001, 21, 1041–1066. [CrossRef]
17. Moazami, S.; Golian, S.; Kavianpour, M.R.; Hong, Y. Comparison of PERSIANN and V7 TRMM Multi-satellite
Precipitation Analysis (TMPA) products with rain gauge data over Iran. Int. J. Remote Sens. 2013, 34,
8156–8171. [CrossRef]
18. Zhao, T.; Yatagai, A. Evaluation of TRMM 3B42 product using a new gauge-based analysis of daily
precipitation over China: Evaluation of TRMM 3B42 Product over China. Int. J. Climatol. 2014, 34, 2749–2762.
[CrossRef]
19. Qiao, L.; Hong, Y.; Chen, S.; Zou, C.B.; Gourley, J.J.; Yong, B. Performance assessment of the successive
Version 6 and Version 7 TMPA products over the climate-transitional zone in the southern Great Plains, USA.
J. Hydrol. 2014, 513, 446–456. [CrossRef]
20. Yang, X.; Yong, B.; Hong, Y.; Chen, S.; Zhang, X. Error analysis of multi-satellite precipitation estimates with
an independent raingauge observation network over a medium-sized humid basin. Hydrol. Sci. J. 2015, 61,
1813–1830. [CrossRef]
21. Zulkafli, Z.; Buytaert, W.; Onof, C.; Manz, B.; Tarnavsky, E.; Lavado, W.; Guyot, J.-L. A Comparative
Performance Analysis of TRMM 3B42 (TMPA) Versions 6 and 7 for Hydrological Applications over
Andean–Amazon River Basins. J. Hydrometeorol. 2014, 15, 581–592. [CrossRef]
22. Yong, B.; Liu, D.; Gourley, J.J.; Tian, Y.; Huffman, G.J.; Ren, L.; Hong, Y. Global view of real-time TRMM
Multi-satellite Precipitation Analysis: Implication to its successor Global Precipitation Measurement mission.
Bull. Am. Meteorol. Soc. 2014. [CrossRef]
23. Katiraie-Boroujerdy, P.-S.; Nasrollahi, N.; Hsu, K.; Sorooshian, S. Evaluation of satellite-based precipitation
estimation over Iran. J. Arid Environ. 2013, 97, 205–219. [CrossRef]
24. Ebert, E.E.; Janowiak, J.E.; Kidd, C. Comparison of Near-Real-Time Precipitation Estimates from Satellite
Observations and Numerical Models. Bull. Am. Meteorol. Soc. 2007, 88, 47–64. [CrossRef]
25. Zhang, X.-X.; Bi, X.-Q.; Kong, X.-H. Observed diurnal cycle of summer precipitation over South Asia and
East Asia based on CMORPH and TRMM satellite data. Atmos. Ocean. Sci. Lett. 2015, 8, 101.
26. Dinku, T.; Ceccato, P.; Cressman, K.; Connor, S.J. Evaluating Detection Skills of Satellite Rainfall Estimates
over Desert Locust Recession Regions. J. Appl. Meteorol. Climatol. 2010, 49, 1322–1332. [CrossRef]
27. Dinku, T.; Ruiz, F.; Connor, S.J.; Ceccato, P. Validation and Intercomparison of Satellite Rainfall Estimates
over Colombia. J. Appl. Meteorol. Climatol. 2010, 49, 1004–1014. [CrossRef]
Sensors 2017, 17, 1413 16 of 17

28. Prakash, S.; Mitra, A.K.; AghaKouchak, A.; Pai, D.S. Error characterization of TRMM Multisatellite
Precipitation Analysis (TMPA-3B42) products over India for different seasons. J. Hydrol. 2015, 529, 1302–1312.
[CrossRef]
29. Gervais, M.; Gyakum, J.R.; Atallah, E.; Tremblay, L.B.; Neale, R.B. How Well Are the Distribution and
Extreme Values of Daily Precipitation over North America Represented in the Community Climate System
Model? A Comparison to Reanalysis, Satellite, and Gridded Station Data. J. Clim. 2014, 27, 5219–5239.
[CrossRef]
30. Asadullah, A.; McINTYRE, N.; Kigobe, M. Evaluation of five satellite products for estimation of rainfall
over Uganda/Evaluation de cinq produits satellitaires pour l’estimation des précipitations en Ouganda.
Hydrol. Sci. J. 2008, 53, 1137–1150. [CrossRef]
31. Ghajarnia, N.; Liaghat, A.; Daneshkar Arasteh, P. Comparison and evaluation of high resolution precipitation
estimation products in Urmia Basin-Iran. Atmos. Res. 2015, 158–159, 50–65. [CrossRef]
32. Ajaaj, A.A.; Mishra, A.K.; Khan, A.A. Comparison of BIAS correction techniques for GPCC rainfall data in
semi-arid climate. Stoch. Environ. Res. Risk Assess. 2015, 30, 1659–1675. [CrossRef]
33. Lenderink, G.; Buishand, A.; van Deursen, W. Estimates of future discharges of the river Rhine using two
scenario methodologies: Direct versus delta approach. Hydrol. Earth Syst. Sci. 2007, 11, 1145–1159. [CrossRef]
34. Boushaki, F.I.; Hsu, K.-L.; Sorooshian, S.; Park, G.-H.; Mahani, S.; Shi, W. Bias Adjustment of Satellite
Precipitation Estimation Using Ground-Based Measurement: A Case Study Evaluation over the Southwestern
United States. J. Hydrometeorol. 2009, 10, 1231–1242. [CrossRef]
35. Lin, A.; Wang, X.L. An algorithm for blending multiple satellite precipitation estimates with in situ
precipitation measurements in Canada: Blended analysis of precipitation. J. Geophys. Res. Atmos. 2011, 116.
[CrossRef]
36. Tesfagiorgis, K.; Mahani, S.E.; Krakauer, N.Y.; Khanbilvardi, R. Bias correction of satellite rainfall estimates
using a radar-gauge product—A case study in Oklahoma (USA). Hydrol. Earth Syst. Sci. 2011, 15, 2631–2647.
37. Teutschbein, C.; Seibert, J. Is bias correction of regional climate model (RCM) simulations possible for
non-stationary conditions? Hydrol. Earth Syst. Sci. 2013, 17, 5061–5077. [CrossRef]
38. Schmidli, J.; Frei, C.; Vidale, P.L. Downscaling from GCM precipitation: A benchmark for dynamical and
statistical downscaling methods. Int. J. Climatol. 2006, 26, 679–689. [CrossRef]
39. Leander, R.; Buishand, T.A.; van den Hurk, B.J.J.M.; de Wit, M.J.M. Estimated changes in flood quantiles of
the river Meuse from resampling of regional climate model output. J. Hydrol. 2008, 351, 331–343. [CrossRef]
40. Leander, R.; Buishand, T.A. Resampling of regional climate model output for the simulation of extreme river
flows. J. Hydrol. 2007, 332, 487–496. [CrossRef]
41. Teutschbein, C.; Seibert, J. Bias correction of regional climate model simulations for hydrological
climate-change impact studies: Review and evaluation of different methods. J. Hydrol. 2012, 456–457,
12–29. [CrossRef]
42. Block, P.J.; Souza Filho, F.A.; Sun, L.; Kwon, H.-H. A Streamflow Forecasting Framework using Multiple
Climate and Hydrological Models. J. Am. Water Resour. Assoc. 2009, 45, 828–843. [CrossRef]
43. Ines, A.V.M.; Hansen, J.W. Bias correction of daily GCM rainfall for crop simulation studies.
Agric. For. Meteorol. 2006, 138, 44–53. [CrossRef]
44. Johnson, F.; Sharma, A. Accounting for interannual variability: A comparison of options for water resources
climate change impact assessments: Accounting for interannual variability. Water Resour. Res. 2011, 47.
[CrossRef]
45. McGinnis, S.; Nychka, D.; Mearns, L.O. A new distribution mapping technique for climate model bias
correction. In Machine Learning and Data Mining Approaches to Climate Science: Proceedings of the Fourth
International Workshop on Climate Informatics; Lakshmanan, V., Gilleland, E., McGovern, A., Tingley, M., Eds.;
Springer: Boulder, CO, USA, 2015.
46. Piani, C.; Haerter, J.O.; Coppola, E. Statistical bias correction for daily precipitation in regional climate
models over Europe. Theor. Appl. Climatol. 2010, 99, 187–192. [CrossRef]
47. Rajczak, J.; Kotlarski, S.; Schär, C. Does Quantile Mapping of Simulated Precipitation Correct for Biases in
Transition Probabilities and Spell Lengths? J. Clim. 2016, 29, 1605–1615. [CrossRef]
48. Sun, F.; Roderick, M.L.; Lim, W.H.; Farquhar, G.D. Hydroclimatic projections for the Murray-Darling Basin
based on an ensemble derived from Intergovernmental Panel on Climate Change AR4 climate models:
Hydroclimatic Projections for the MDB. Water Resour. Res. 2011, 47. [CrossRef]
Sensors 2017, 17, 1413 17 of 17

49. Wood, A.W.; Leung, L.R.; Sridhar, V.; Lettenmaier, D.P. Hydrologic Implications of Dynamical and Statistical
Approaches to Downscaling Climate Model Outputs. Clim. Chang. 2004, 62, 189–216. [CrossRef]
50. Themeßl, M.J.; Gobiet, A.; Heinrich, G. Empirical-statistical downscaling and error correction of regional
climate models and its impact on the climate change signal. Clim. Chang. 2012, 112, 449–468. [CrossRef]
51. Bennett, J.C.; Grose, M.R.; Corney, S.P.; White, C.J.; Holz, G.K.; Katzfey, J.J.; Post, D.A.; Bindoff, N.L.
Performance of an empirical bias-correction of a high-resolution climate dataset: Empirical bias-correction of
a high-resolution climate dataset. Int. J. Climatol. 2014, 34, 2189–2204. [CrossRef]
52. Yang, Z.; Hsu, K.; Sorooshian, S.; Xu, X.; Braithwaite, D.; Verbist, K.M. J. Bias adjustment of satellite-based
precipitation estimation using gauge observations: A case study in Chile: Satellite Precipitation Bias
Adjustment. J. Geophys. Res. Atmos. 2016, 121, 3790–3806. [CrossRef]
53. Hammond, D.S. Tropical Forests of the Guiana Shield: Ancient Forests in a Modern World; CABI: Oxfordshire,
UK, 2005.
54. Bovolo, C.I.; Pereira, R.; Parkin, G.; Kilsby, C.; Wagner, T. Fine-scale regional climate patterns in the Guianas,
tropical South America, based on observations and reanalysis data. Int. J. Climatol. 2012, 32, 1665–1689.
[CrossRef]
55. Hidroweb. Available online: http://www.snirh.gov.br/hidroweb/ (accessed on 10 January 2014).
56. Yong, B.; Ren, L.; Hong, Y.; Gourley, J.J.; Tian, Y.; Huffman, G.J.; Chen, X.; Wang, W.; Wen, Y. First evaluation
of the climatological calibration algorithm in the real-time TMPA precipitation estimates over two basins at
high and low latitudes: Validation of crutial algorithmic upgrade in NASA TMPA. Water Resour. Res. 2013,
49, 2461–2472. [CrossRef]
57. Yong, B.; Hong, Y.; Ren, L.-L.; Gourley, J.J.; Huffman, G.J.; Chen, X.; Wang, W.; Khan, S.I. Assessment
of evolving TRMM-based multisatellite real-time precipitation estimation methods and their impacts on
hydrologic prediction in a high latitude basin. J. Geophys. Res. 2012, 117. [CrossRef]
58. NASA. Available online: ftp://disc2.nascom.nasa.gov/data/TRMM/Gridded/3B42_V7/ (accessed on
14 September 2014).
59. Ward, J.H. Hierarchical Grouping to Optimize an Objective Function. J. Am. Stat. Assoc. 1963, 58, 236–244.
[CrossRef]
60. Gordon, A.D. A Review of Hierarchical Classification. J. R. Stat. Soc. 1987, 150, 119–137. [CrossRef]
61. Piani, C.; Weedon, G.P.; Best, M.; Gomes, S.M.; Viterbo, P.; Hagemann, S.; Haerter, J.O. Statistical bias
correction of global simulated daily precipitation and temperature for the application of hydrological models.
J. Hydrol. 2010, 395, 199–215. [CrossRef]
62. Gudmundsson, L.; Bremnes, J.B.; Haugen, J.E.; Engen-Skaugen, T. Technical Note: Downscaling RCM
precipitation to the station scale using statistical transformations—A comparison of methods. Hydrol. Earth
Syst. Sci. 2012, 16, 3383–3390. [CrossRef]
63. Kim, K.B.; Bray, M.; Han, D. An improved bias correction scheme based on comparative precipitation
characteristics: Bias correction based on precipitation characteristics. Hydrol. Process. 2015, 29, 2258–2266.
[CrossRef]
64. Lafon, T.; Dadson, S.; Buys, G.; Prudhomme, C. Bias correction of daily precipitation simulated by a regional
climate model: A comparison of methods. Int. J. Climatol. 2013, 33, 1367–1381. [CrossRef]
65. Zambrano-Bigiarini, M.; Nauditt, A.; Birkel, C.; Verbist, K.; Ribbe, L. Temporal and spatial evaluation
of satellite-based rainfall estimates across the complex topographical and climatic gradients of Chile.
Hydrol. Earth Syst. Sci. 2017, 21, 1295–1320. [CrossRef]

© 2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).